Example usage for org.apache.hadoop.mapreduce MRJobConfig MAP_OUTPUT_COMPRESS_CODEC

Introduction

In this page you can find the example usage for org.apache.hadoop.mapreduce MRJobConfig MAP_OUTPUT_COMPRESS_CODEC.

Prototype

String MAP_OUTPUT_COMPRESS_CODEC

To view the source code for org.apache.hadoop.mapreduce MRJobConfig MAP_OUTPUT_COMPRESS_CODEC.

Click Source Link

Usage

From source file:com.cloudera.oryx.computation.common.JobStep.java

License:Open Source License

/**
 * Creates a new {@link MRPipeline} instance that contains common configuration
 * settings.//from   w ww  . j  a  v  a 2 s .  co  m
 *
 * @return a new {@link MRPipeline} instance, suitably configured
 */
protected final MRPipeline createBasicPipeline(Class<?> jarClass) throws IOException {
    Configuration conf = OryxConfiguration.get(getConf());

    conf.setBoolean(MRJobConfig.MAP_OUTPUT_COMPRESS, true);
    conf.setClass(MRJobConfig.MAP_OUTPUT_COMPRESS_CODEC, SnappyCodec.class, CompressionCodec.class);

    conf.setBoolean("mapred.output.compress", true);
    conf.set("mapred.output.compression.type", "BLOCK");
    conf.setClass("mapred.output.compression.codec", SnappyCodec.class, CompressionCodec.class);
    // Set old-style equivalents for Avro/Crunch's benefit
    conf.set("avro.output.codec", "snappy");

    conf.setBoolean(MRJobConfig.MAP_SPECULATIVE, true);
    conf.setBoolean(MRJobConfig.REDUCE_SPECULATIVE, true);
    conf.setBoolean(TTConfig.TT_OUTOFBAND_HEARBEAT, true);
    conf.setInt(MRJobConfig.JVM_NUMTASKS_TORUN, -1);

    //conf.setBoolean("crunch.disable.deep.copy", true);
    // Giving one mapper a lot of data can cause issues in some stages, so default to disable this
    conf.setBoolean("crunch.disable.combine.file", true);

    Config appConfig = ConfigUtils.getDefaultConfig();

    conf.set("crunch.tmp.dir", appConfig.getString("computation-layer.tmp-dir"));

    int mapMemoryMB = appConfig.getInt("computation-layer.mapper-memory-mb");
    log.info("Mapper memory: {}", mapMemoryMB);
    int mapHeapMB = (int) (mapMemoryMB / 1.3); // Matches Hadoop's default
    log.info("Mappers have {}MB heap and can access {}MB RAM", mapHeapMB, mapMemoryMB);
    if (conf.get(MRJobConfig.MAP_JAVA_OPTS) != null) {
        log.info("Overriding previous setting of {}, which was '{}'", MRJobConfig.MAP_JAVA_OPTS,
                conf.get(MRJobConfig.MAP_JAVA_OPTS));
    }
    conf.set(MRJobConfig.MAP_JAVA_OPTS,
            "-Xmx" + mapHeapMB + "m -XX:+UseCompressedOops -XX:+UseParallelGC -XX:+UseParallelOldGC");
    log.info("Set {} to '{}'", MRJobConfig.MAP_JAVA_OPTS, conf.get(MRJobConfig.MAP_JAVA_OPTS));
    // See comment below on CM
    conf.setInt("mapreduce.map.java.opts.max.heap", mapHeapMB);

    int reduceMemoryMB = appConfig.getInt("computation-layer.reducer-memory-mb");
    log.info("Reducer memory: {}", reduceMemoryMB);
    if (isHighMemoryStep()) {
        reduceMemoryMB *= appConfig.getInt("computation-layer.worker-high-memory-factor");
        log.info("Increasing {} to {} for high-memory step", MRJobConfig.REDUCE_MEMORY_MB, reduceMemoryMB);
    }
    conf.setInt(MRJobConfig.REDUCE_MEMORY_MB, reduceMemoryMB);

    int reduceHeapMB = (int) (reduceMemoryMB / 1.3); // Matches Hadoop's default
    log.info("Reducers have {}MB heap and can access {}MB RAM", reduceHeapMB, reduceMemoryMB);
    if (conf.get(MRJobConfig.REDUCE_JAVA_OPTS) != null) {
        log.info("Overriding previous setting of {}, which was '{}'", MRJobConfig.REDUCE_JAVA_OPTS,
                conf.get(MRJobConfig.REDUCE_JAVA_OPTS));
    }
    conf.set(MRJobConfig.REDUCE_JAVA_OPTS,
            "-Xmx" + reduceHeapMB + "m -XX:+UseCompressedOops -XX:+UseParallelGC -XX:+UseParallelOldGC");
    log.info("Set {} to '{}'", MRJobConfig.REDUCE_JAVA_OPTS, conf.get(MRJobConfig.REDUCE_JAVA_OPTS));
    // I see this in CM but not in Hadoop docs; probably won't hurt as it's supposed to result in
    // -Xmx appended to opts above, which is at worst redundant
    conf.setInt("mapreduce.reduce.java.opts.max.heap", reduceHeapMB);

    conf.setInt("yarn.scheduler.capacity.minimum-allocation-mb", 128);
    conf.setInt("yarn.app.mapreduce.am.resource.mb", 384);

    // Pass total config state
    conf.set(CONFIG_SERIALIZATION_KEY, ConfigUtils.getDefaultConfig().root().render());

    // Make sure to set any args to conf above this line!

    setConf(conf);

    Job job = Job.getInstance(conf);

    // Basic File IO settings
    FileInputFormat.setMaxInputSplitSize(job, 1L << 28); // ~268MB
    SequenceFileOutputFormat.setOutputCompressionType(job, SequenceFile.CompressionType.BLOCK);
    FileOutputFormat.setCompressOutput(job, true);
    FileOutputFormat.setOutputCompressorClass(job, SnappyCodec.class);

    log.info("Created pipeline configuration {}", job.getConfiguration());

    return new MRPipeline(jarClass, getCustomJobName(), job.getConfiguration());
}

---

From source file:com.cloudera.oryx.computation.common.JobStep.java

License:Open Source License

protected final Target compressedTextOutput(Configuration conf, String outputPathKey) {
    // The way this is used, it doesn't seem like we can just set the object in getConf(). Need
    // to set the copy in the MRPipeline directly?
    conf.setClass(FileOutputFormat.COMPRESS_CODEC, GzipCodec.class, CompressionCodec.class);
    conf.setClass(MRJobConfig.MAP_OUTPUT_COMPRESS_CODEC, SnappyCodec.class, CompressionCodec.class);
    return To.textFile(Namespaces.toPath(outputPathKey));
}

---

From source file:com.msd.gin.halyard.tools.HalyardBulkLoad.java

License:Apache License

@Override
public int run(String[] args) throws Exception {
    if (args.length != 3) {
        System.err.println("Usage: bulkload [-D" + MRJobConfig.QUEUE_NAME + "=proofofconcepts] [-D"
                + SKIP_INVALID_PROPERTY + "=true] [-D" + SPLIT_BITS_PROPERTY + "=8] [-D"
                + DEFAULT_CONTEXT_PROPERTY + "=http://new_context] [-D" + OVERRIDE_CONTEXT_PROPERTY
                + "=true] <input_path(s)> <output_path> <table_name>");
        return -1;
    }//w w w.j a  va2  s .  co m
    TableMapReduceUtil.addDependencyJars(getConf(), NTriplesUtil.class, Rio.class, AbstractRDFHandler.class,
            RDFFormat.class, RDFParser.class);
    HBaseConfiguration.addHbaseResources(getConf());
    if (SnappyCodec.isNativeCodeLoaded()) {
        getConf().setBoolean(MRJobConfig.MAP_OUTPUT_COMPRESS, true);
        getConf().setClass(MRJobConfig.MAP_OUTPUT_COMPRESS_CODEC, SnappyCodec.class, CompressionCodec.class);
    }
    getConf().setDouble(MRJobConfig.COMPLETED_MAPS_FOR_REDUCE_SLOWSTART, 1.0);
    getConf().setLong(MRJobConfig.TASK_TIMEOUT, 3600000l);
    getConf().setInt(MRJobConfig.IO_SORT_FACTOR, 100);
    getConf().setInt(MRJobConfig.IO_SORT_MB, 1000);
    getConf().setInt(FileInputFormat.SPLIT_MAXSIZE, 1000000000);
    getConf().setInt(LoadIncrementalHFiles.MAX_FILES_PER_REGION_PER_FAMILY, 2048);
    Job job = Job.getInstance(getConf(), "HalyardBulkLoad -> " + args[1] + " -> " + args[2]);
    job.setJarByClass(HalyardBulkLoad.class);
    job.setMapperClass(RDFMapper.class);
    job.setMapOutputKeyClass(ImmutableBytesWritable.class);
    job.setMapOutputValueClass(KeyValue.class);
    job.setInputFormatClass(RioFileInputFormat.class);
    job.setSpeculativeExecution(false);
    job.setReduceSpeculativeExecution(false);
    Map<String, Integer> contextSplitsMap = new HashMap<>();
    for (Map.Entry<String, String> me : getConf().getValByRegex(CONTEXT_SPLIT_REGEXP).entrySet()) {
        int splits = Integer.parseInt(me.getKey().substring(me.getKey().lastIndexOf('.') + 1));
        StringTokenizer stk = new StringTokenizer(me.getValue(), ",");
        while (stk.hasMoreTokens()) {
            contextSplitsMap.put(stk.nextToken(), splits);
        }
    }
    try (HTable hTable = HalyardTableUtils.getTable(getConf(), args[2], true,
            getConf().getInt(SPLIT_BITS_PROPERTY, 3), contextSplitsMap)) {
        HFileOutputFormat2.configureIncrementalLoad(job, hTable.getTableDescriptor(),
                hTable.getRegionLocator());
        FileInputFormat.setInputDirRecursive(job, true);
        FileInputFormat.setInputPaths(job, args[0]);
        FileOutputFormat.setOutputPath(job, new Path(args[1]));
        TableMapReduceUtil.addDependencyJars(job);
        TableMapReduceUtil.initCredentials(job);
        if (job.waitForCompletion(true)) {
            new LoadIncrementalHFiles(getConf()).doBulkLoad(new Path(args[1]), hTable);
            LOG.info("Bulk Load Completed..");
            return 0;
        }
    }
    return -1;
}

---

From source file:com.msd.gin.halyard.tools.HalyardBulkUpdate.java

License:Apache License

@Override
public int run(String[] args) throws Exception {
    if (args.length != 3) {
        System.err.println("Usage: bulkupdate [-D" + MRJobConfig.QUEUE_NAME + "=proofofconcepts] [-D"
                + DEFAULT_CONTEXT_PROPERTY + "=http://new_context] [-D" + OVERRIDE_CONTEXT_PROPERTY
                + "=true] <input_file_with_SPARQL_queries> <output_path> <table_name>");
        return -1;
    }/*from   ww w.  j ava 2 s.  c  o  m*/
    TableMapReduceUtil.addDependencyJars(getConf(), NTriplesUtil.class, Rio.class, RDFFormat.class,
            RDFParser.class);
    HBaseConfiguration.addHbaseResources(getConf());
    if (SnappyCodec.isNativeCodeLoaded()) {
        getConf().setBoolean(MRJobConfig.MAP_OUTPUT_COMPRESS, true);
        getConf().setClass(MRJobConfig.MAP_OUTPUT_COMPRESS_CODEC, SnappyCodec.class, CompressionCodec.class);
    }
    getConf().setDouble(MRJobConfig.COMPLETED_MAPS_FOR_REDUCE_SLOWSTART, 1.0);
    getConf().setLong(MRJobConfig.TASK_TIMEOUT, 3600000l);
    getConf().setInt(MRJobConfig.IO_SORT_FACTOR, 100);
    getConf().setInt(MRJobConfig.IO_SORT_MB, 1000);
    getConf().setInt(FileInputFormat.SPLIT_MAXSIZE, 1000000000);
    getConf().setInt(LoadIncrementalHFiles.MAX_FILES_PER_REGION_PER_FAMILY, 2048);
    getConf().setStrings(TABLE_NAME_PROPERTY, args[2]);
    Job job = Job.getInstance(getConf(), "HalyardBulkUpdate -> " + args[1] + " -> " + args[2]);
    NLineInputFormat.setNumLinesPerSplit(job, 1);
    job.setJarByClass(HalyardBulkUpdate.class);
    job.setMapperClass(SPARQLMapper.class);
    job.setMapOutputKeyClass(ImmutableBytesWritable.class);
    job.setMapOutputValueClass(KeyValue.class);
    job.setInputFormatClass(NLineInputFormat.class);
    job.setSpeculativeExecution(false);
    job.setReduceSpeculativeExecution(false);
    try (HTable hTable = HalyardTableUtils.getTable(getConf(), args[2], false, 0, null)) {
        HFileOutputFormat2.configureIncrementalLoad(job, hTable.getTableDescriptor(),
                hTable.getRegionLocator());
        FileInputFormat.setInputPaths(job, args[0]);
        FileOutputFormat.setOutputPath(job, new Path(args[1]));
        TableMapReduceUtil.addDependencyJars(job);
        TableMapReduceUtil.initCredentials(job);
        if (job.waitForCompletion(true)) {
            new LoadIncrementalHFiles(getConf()).doBulkLoad(new Path(args[1]), hTable);
            LOG.info("Bulk Update Completed..");
            return 0;
        }
    }
    return -1;
}

---

From source file:com.msd.gin.halyard.tools.HalyardHiveLoad.java

License:Apache License

@Override
public int run(String[] args) throws Exception {
    if (args.length != 3) {
        System.err.println("Usage: hiveload -D" + RDF_MIME_TYPE_PROPERTY + "='application/ld+json' [-D"
                + MRJobConfig.QUEUE_NAME + "=proofofconcepts] [-D" + HIVE_DATA_COLUMN_INDEX_PROPERTY + "=3] [-D"
                + BASE_URI_PROPERTY + "='http://my_base_uri/'] [-D" + HalyardBulkLoad.SPLIT_BITS_PROPERTY
                + "=8] [-D" + HalyardBulkLoad.DEFAULT_CONTEXT_PROPERTY + "=http://new_context] [-D"
                + HalyardBulkLoad.OVERRIDE_CONTEXT_PROPERTY
                + "=true] <hive_table_name> <output_path> <hbase_table_name>");
        return -1;
    }//from w w  w.ja v  a 2 s .c o  m
    TableMapReduceUtil.addDependencyJars(getConf(), NTriplesUtil.class, Rio.class, AbstractRDFHandler.class,
            RDFFormat.class, RDFParser.class);
    HBaseConfiguration.addHbaseResources(getConf());
    if (SnappyCodec.isNativeCodeLoaded()) {
        getConf().setBoolean(MRJobConfig.MAP_OUTPUT_COMPRESS, true);
        getConf().setClass(MRJobConfig.MAP_OUTPUT_COMPRESS_CODEC, SnappyCodec.class, CompressionCodec.class);
    }
    getConf().setDouble(MRJobConfig.COMPLETED_MAPS_FOR_REDUCE_SLOWSTART, 1.0);
    getConf().setLong(MRJobConfig.TASK_TIMEOUT, 3600000l);
    getConf().setInt(MRJobConfig.IO_SORT_FACTOR, 100);
    getConf().setInt(MRJobConfig.IO_SORT_MB, 1000);
    getConf().setInt(FileInputFormat.SPLIT_MAXSIZE, 1000000000);
    getConf().setInt(LoadIncrementalHFiles.MAX_FILES_PER_REGION_PER_FAMILY, 2048);
    Job job = Job.getInstance(getConf(), "HalyardHiveLoad -> " + args[1] + " -> " + args[2]);
    int i = args[0].indexOf('.');
    HCatInputFormat.setInput(job, i > 0 ? args[0].substring(0, i) : null, args[0].substring(i + 1));
    job.setJarByClass(HalyardHiveLoad.class);
    job.setMapperClass(HiveMapper.class);
    job.setMapOutputKeyClass(ImmutableBytesWritable.class);
    job.setMapOutputValueClass(KeyValue.class);
    job.setInputFormatClass(HCatInputFormat.class);
    job.setSpeculativeExecution(false);
    job.setReduceSpeculativeExecution(false);
    Map<String, Integer> contextSplitsMap = new HashMap<>();
    for (Map.Entry<String, String> me : getConf().getValByRegex(HalyardBulkLoad.CONTEXT_SPLIT_REGEXP)
            .entrySet()) {
        int splits = Integer.parseInt(me.getKey().substring(me.getKey().lastIndexOf('.') + 1));
        StringTokenizer stk = new StringTokenizer(me.getValue(), ",");
        while (stk.hasMoreTokens()) {
            contextSplitsMap.put(stk.nextToken(), splits);
        }
    }
    try (HTable hTable = HalyardTableUtils.getTable(getConf(), args[2], true,
            getConf().getInt(HalyardBulkLoad.SPLIT_BITS_PROPERTY, 3), contextSplitsMap)) {
        HFileOutputFormat2.configureIncrementalLoad(job, hTable.getTableDescriptor(),
                hTable.getRegionLocator());
        FileInputFormat.setInputDirRecursive(job, true);
        FileInputFormat.setInputPaths(job, args[0]);
        FileOutputFormat.setOutputPath(job, new Path(args[1]));
        TableMapReduceUtil.addDependencyJars(job);
        TableMapReduceUtil.initCredentials(job);
        if (job.waitForCompletion(true)) {
            new LoadIncrementalHFiles(getConf()).doBulkLoad(new Path(args[1]), hTable);
            LOG.info("Bulk Load Completed..");
            return 0;
        }
    }
    return -1;
}

---